SPC
Statistical process control asks one question: is this process stable, or is something new happening? Control limits come from the process itself. If you pick them to match the specification, you have built a chart that agrees with you.
The quality practice.
Three projects written up properly: what the problem actually was, which framework I reached for, what I built, and what changed afterwards. Followed by the methods themselves, in plain language, for anyone who has to run them.
Case one · Process improvement
Cutting defects on a high-volume line with 8D and SPC.
A high-volume assembly line was leaking defects into final inspection and, occasionally, to the customer. Rework and scrap were climbing and the customer quality scorecard was at risk.
~40%Defect reduction
Real-timeOut-of-control alerts
AuditReady SPC documentation
8DCross-functional method
The problem underneath the problem
There were four plausible causes at once: material, machine, method and environment. Nobody could rule any of them out, because there was no real-time visibility into whether the process was stable in the first place. Quality action was reactive by construction, and the documentation would not have survived an IATF or ISO look at corrective action.
What I did
I ran it as an 8D, which forces a cross-functional team, a containment step before anyone argues about root cause, and a verification step before anyone declares victory. 5-Why inside the 8D drove from symptom down to the process and system causes rather than stopping at the operator.
Then I put the line under statistical process control. Control charts for the key characteristics, with limits calculated from historical data rather than picked to look reasonable, and Power BI dashboards that alerted on out-of-control conditions rather than waiting for the daily report. Operators were trained on what to actually do when a point went out, because a control chart nobody responds to is decoration.
What changed
Defects fell by roughly 40%. More importantly, containment got faster, because the alert arrived while the process was still drifting instead of after a bad lot had shipped. And the SPC records became the audit evidence, so audit preparation stopped being a scramble.
Run the numbers yourself with the process capability and control-limit calculators.
Case two · Analytics
One place to see incoming, in-process and customer quality.
Quality data lived in Excel workbooks, legacy databases, and occasionally paper. Producing a picture for leadership meant manual pulls, copy-paste and reconciliation, and the answer was usually stale by the time it arrived.
~60%Less manual reporting
Same-dayVisibility for management
Plant → partDrill-down depth
OneDefinition per metric
The real failure was definitional
The reporting time was the visible symptom. The underlying problem was that "incoming defect rate" meant something slightly different in each workbook, so two people could both be right and disagree. Any dashboard built on top of that would just have industrialised the disagreement.
What I built
A proper ETL layer: extract from the source systems, transform and standardise in SQL and Power Query, load into a schema Power BI could consume. That is where the single definition of each metric got fixed, once.
On top of it, a star schema - facts for inspection results and defect counts, dimensions for time, plant, part and supplier - so leadership could slice by any dimension without anyone duplicating the logic in a new measure. KPIs were DAX measures against that model, refreshed on a schedule rather than on request.
What changed
Manual reporting dropped by roughly 60%, which mattered less than the second-order effect: the quality engineers got that time back for analysis instead of assembly. Standups ran off numbers everyone agreed on, and a question could be answered by drilling down rather than by commissioning a new report.
Case three · Quality systems
A CAPA workflow that could answer "where is the evidence?"
Nonconformances were in one system, corrective actions in another, and the evidence that closed them was in email and shared folders. Every audit began with a scramble.
ZeroMajor findings, follow-up audit
Clause 10.2ISO 9001:2015 traceability
6-stageStandard CAPA workflow
One packSupplier scorecards
Why closure dates kept slipping
Ownership was ambiguous at the moment a CAPA was opened, so nobody was late until everybody was. Duplicate and outdated procedure versions meant two people could implement a correction against different documents and both believe they were done.
What I standardised
A six-stage workflow with required fields at each gate: Open → Root cause → Action plan → Implementation → Verification → Close. Ownership assigned at open, escalation when overdue. Each stage designed so it could be demonstrated to an auditor rather than merely recorded.
Then traceability: every CAPA linked to its originating nonconformance or finding, to the relevant QMS procedures, and to the evidence that closed it. That link set is what makes an audit pack possible, and it is the part teams usually skip.
What changed
The follow-up audit closed with zero major findings. Closure got faster because ownership was unambiguous from the first stage, and supplier audit scorecards ended up in one place instead of five.
Methods
The toolkit, in plain language.
What each method is actually for, and the failure mode it is usually reached for too late to prevent.
Process capability
Cp, Cpk, Pp and Ppk compare what the process can do against what the drawing demands. Cpk without stability is a number describing a process that is not behaving the same way twice.
Gage R&R
Before trusting any measurement, find out how much of the variation is the measurement system rather than the part. This is the study that tells you whether the rest of your data means anything.
AQL sampling
Acceptance sampling sets how many units to inspect and how many defects are tolerable. The plan has to be tied to risk and to the customer's specification, or it is indefensible the moment it is questioned.
8D and 5-Why
A disciplined route from symptom to system cause, with containment before analysis and verification before closure. The discipline is the point; the form is not.
FMEA
Failure mode and effects analysis works forwards from what could go wrong, before it has. Its value is entirely in being done early, which is exactly when nobody wants to spend the time.
CAPA
Corrective and preventive action, traceable from the finding through the fix to the evidence. ISO 9001:2015 Clause 10.2 is the shape of it.
ISO 9001 and 19011
9001 sets what the quality management system must do; 19011 sets how to audit it. Most of the work is keeping documents matched to what the floor actually does.
PPAP
Production part approval: proving a part and its process are ready before volume starts, so the first shipment is not the first test.
Calculators for most of the above
AQL sampling plans, process capability, Gage R&R and control limits, running entirely in your browser. Nothing is uploaded and there is no sign-up.
Next
Need this run on your process?
Open to quality and data engineering roles, and to consulting engagements.